This is an article that I co-authored with my friend and collaborator Kate Crawford, who directs the AI Now Institute at NYU. In the article we take a look at some of the bad assumptions and bad politics built into the architecture of the training data used in AI systems.